Arnaldo Carvalho de Melo | 8feaf0c | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 1 | /* |
| 2 | * INET An implementation of the TCP/IP protocol suite for the LINUX |
| 3 | * operating system. INET is implemented using the BSD Socket |
| 4 | * interface as the means of communication with the user level. |
| 5 | * |
| 6 | * Definitions for a generic INET TIMEWAIT sock |
| 7 | * |
| 8 | * From code originally in net/tcp.h |
| 9 | * |
| 10 | * This program is free software; you can redistribute it and/or |
| 11 | * modify it under the terms of the GNU General Public License |
| 12 | * as published by the Free Software Foundation; either version |
| 13 | * 2 of the License, or (at your option) any later version. |
| 14 | */ |
| 15 | #ifndef _INET_TIMEWAIT_SOCK_ |
| 16 | #define _INET_TIMEWAIT_SOCK_ |
| 17 | |
Arnaldo Carvalho de Melo | 8feaf0c | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 18 | |
| 19 | #include <linux/list.h> |
Arnaldo Carvalho de Melo | eeb2b85 | 2005-10-10 21:25:23 -0700 | [diff] [blame] | 20 | #include <linux/module.h> |
Arnaldo Carvalho de Melo | 295ff7e | 2005-08-09 20:44:40 -0700 | [diff] [blame] | 21 | #include <linux/timer.h> |
Arnaldo Carvalho de Melo | 8feaf0c | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 22 | #include <linux/types.h> |
Arnaldo Carvalho de Melo | 295ff7e | 2005-08-09 20:44:40 -0700 | [diff] [blame] | 23 | #include <linux/workqueue.h> |
Arnaldo Carvalho de Melo | 8feaf0c | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 24 | |
Arnaldo Carvalho de Melo | 14c8502 | 2005-12-27 02:43:12 -0200 | [diff] [blame] | 25 | #include <net/inet_sock.h> |
Arnaldo Carvalho de Melo | 8feaf0c | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 26 | #include <net/sock.h> |
| 27 | #include <net/tcp_states.h> |
Arnaldo Carvalho de Melo | 6d6ee43 | 2005-12-13 23:25:19 -0800 | [diff] [blame] | 28 | #include <net/timewait_sock.h> |
Arnaldo Carvalho de Melo | 8feaf0c | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 29 | |
| 30 | #include <asm/atomic.h> |
| 31 | |
Arnaldo Carvalho de Melo | 295ff7e | 2005-08-09 20:44:40 -0700 | [diff] [blame] | 32 | struct inet_hashinfo; |
| 33 | |
| 34 | #define INET_TWDR_RECYCLE_SLOTS_LOG 5 |
| 35 | #define INET_TWDR_RECYCLE_SLOTS (1 << INET_TWDR_RECYCLE_SLOTS_LOG) |
| 36 | |
| 37 | /* |
| 38 | * If time > 4sec, it is "slow" path, no recycling is required, |
| 39 | * so that we select tick to get range about 4 seconds. |
| 40 | */ |
| 41 | #if HZ <= 16 || HZ > 4096 |
| 42 | # error Unsupported: HZ <= 16 or HZ > 4096 |
| 43 | #elif HZ <= 32 |
| 44 | # define INET_TWDR_RECYCLE_TICK (5 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG) |
| 45 | #elif HZ <= 64 |
| 46 | # define INET_TWDR_RECYCLE_TICK (6 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG) |
| 47 | #elif HZ <= 128 |
| 48 | # define INET_TWDR_RECYCLE_TICK (7 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG) |
| 49 | #elif HZ <= 256 |
| 50 | # define INET_TWDR_RECYCLE_TICK (8 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG) |
| 51 | #elif HZ <= 512 |
| 52 | # define INET_TWDR_RECYCLE_TICK (9 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG) |
| 53 | #elif HZ <= 1024 |
| 54 | # define INET_TWDR_RECYCLE_TICK (10 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG) |
| 55 | #elif HZ <= 2048 |
| 56 | # define INET_TWDR_RECYCLE_TICK (11 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG) |
| 57 | #else |
| 58 | # define INET_TWDR_RECYCLE_TICK (12 + 2 - INET_TWDR_RECYCLE_SLOTS_LOG) |
| 59 | #endif |
| 60 | |
| 61 | /* TIME_WAIT reaping mechanism. */ |
| 62 | #define INET_TWDR_TWKILL_SLOTS 8 /* Please keep this a power of 2. */ |
| 63 | |
| 64 | #define INET_TWDR_TWKILL_QUOTA 100 |
| 65 | |
| 66 | struct inet_timewait_death_row { |
| 67 | /* Short-time timewait calendar */ |
| 68 | int twcal_hand; |
| 69 | int twcal_jiffie; |
| 70 | struct timer_list twcal_timer; |
| 71 | struct hlist_head twcal_row[INET_TWDR_RECYCLE_SLOTS]; |
| 72 | |
| 73 | spinlock_t death_lock; |
| 74 | int tw_count; |
| 75 | int period; |
| 76 | u32 thread_slots; |
| 77 | struct work_struct twkill_work; |
| 78 | struct timer_list tw_timer; |
| 79 | int slot; |
| 80 | struct hlist_head cells[INET_TWDR_TWKILL_SLOTS]; |
| 81 | struct inet_hashinfo *hashinfo; |
| 82 | int sysctl_tw_recycle; |
| 83 | int sysctl_max_tw_buckets; |
| 84 | }; |
| 85 | |
Arnaldo Carvalho de Melo | 696ab2d | 2005-08-09 20:45:03 -0700 | [diff] [blame] | 86 | extern void inet_twdr_hangman(unsigned long data); |
| 87 | extern void inet_twdr_twkill_work(void *data); |
| 88 | extern void inet_twdr_twcal_tick(unsigned long data); |
| 89 | |
Arnaldo Carvalho de Melo | 8feaf0c | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 90 | #if (BITS_PER_LONG == 64) |
| 91 | #define INET_TIMEWAIT_ADDRCMP_ALIGN_BYTES 8 |
| 92 | #else |
| 93 | #define INET_TIMEWAIT_ADDRCMP_ALIGN_BYTES 4 |
| 94 | #endif |
| 95 | |
| 96 | struct inet_bind_bucket; |
| 97 | |
| 98 | /* |
| 99 | * This is a TIME_WAIT sock. It works around the memory consumption |
| 100 | * problems of sockets in such a state on heavily loaded servers, but |
| 101 | * without violating the protocol specification. |
| 102 | */ |
| 103 | struct inet_timewait_sock { |
| 104 | /* |
| 105 | * Now struct sock also uses sock_common, so please just |
| 106 | * don't add nothing before this first member (__tw_common) --acme |
| 107 | */ |
| 108 | struct sock_common __tw_common; |
| 109 | #define tw_family __tw_common.skc_family |
| 110 | #define tw_state __tw_common.skc_state |
| 111 | #define tw_reuse __tw_common.skc_reuse |
| 112 | #define tw_bound_dev_if __tw_common.skc_bound_dev_if |
| 113 | #define tw_node __tw_common.skc_node |
| 114 | #define tw_bind_node __tw_common.skc_bind_node |
| 115 | #define tw_refcnt __tw_common.skc_refcnt |
Eric Dumazet | 81c3d54 | 2005-10-03 14:13:38 -0700 | [diff] [blame] | 116 | #define tw_hash __tw_common.skc_hash |
Arnaldo Carvalho de Melo | 8feaf0c | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 117 | #define tw_prot __tw_common.skc_prot |
| 118 | volatile unsigned char tw_substate; |
| 119 | /* 3 bits hole, try to pack */ |
| 120 | unsigned char tw_rcv_wscale; |
| 121 | /* Socket demultiplex comparisons on incoming packets. */ |
| 122 | /* these five are in inet_sock */ |
| 123 | __u16 tw_sport; |
| 124 | __u32 tw_daddr __attribute__((aligned(INET_TIMEWAIT_ADDRCMP_ALIGN_BYTES))); |
| 125 | __u32 tw_rcv_saddr; |
| 126 | __u16 tw_dport; |
| 127 | __u16 tw_num; |
| 128 | /* And these are ours. */ |
| 129 | __u8 tw_ipv6only:1; |
Arnaldo Carvalho de Melo | 0fa1a53 | 2005-12-13 23:23:09 -0800 | [diff] [blame] | 130 | /* 15 bits hole, try to pack */ |
| 131 | __u16 tw_ipv6_offset; |
Arnaldo Carvalho de Melo | 8feaf0c | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 132 | int tw_timeout; |
| 133 | unsigned long tw_ttd; |
| 134 | struct inet_bind_bucket *tw_tb; |
| 135 | struct hlist_node tw_death_node; |
| 136 | }; |
| 137 | |
| 138 | static inline void inet_twsk_add_node(struct inet_timewait_sock *tw, |
| 139 | struct hlist_head *list) |
| 140 | { |
| 141 | hlist_add_head(&tw->tw_node, list); |
| 142 | } |
| 143 | |
| 144 | static inline void inet_twsk_add_bind_node(struct inet_timewait_sock *tw, |
| 145 | struct hlist_head *list) |
| 146 | { |
| 147 | hlist_add_head(&tw->tw_bind_node, list); |
| 148 | } |
| 149 | |
| 150 | static inline int inet_twsk_dead_hashed(const struct inet_timewait_sock *tw) |
| 151 | { |
Akinobu Mita | da753be | 2006-04-28 15:21:23 -0700 | [diff] [blame] | 152 | return !hlist_unhashed(&tw->tw_death_node); |
Arnaldo Carvalho de Melo | 8feaf0c | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 153 | } |
| 154 | |
| 155 | static inline void inet_twsk_dead_node_init(struct inet_timewait_sock *tw) |
| 156 | { |
| 157 | tw->tw_death_node.pprev = NULL; |
| 158 | } |
| 159 | |
| 160 | static inline void __inet_twsk_del_dead_node(struct inet_timewait_sock *tw) |
| 161 | { |
| 162 | __hlist_del(&tw->tw_death_node); |
| 163 | inet_twsk_dead_node_init(tw); |
| 164 | } |
| 165 | |
| 166 | static inline int inet_twsk_del_dead_node(struct inet_timewait_sock *tw) |
| 167 | { |
| 168 | if (inet_twsk_dead_hashed(tw)) { |
| 169 | __inet_twsk_del_dead_node(tw); |
| 170 | return 1; |
| 171 | } |
| 172 | return 0; |
| 173 | } |
| 174 | |
| 175 | #define inet_twsk_for_each(tw, node, head) \ |
| 176 | hlist_for_each_entry(tw, node, head, tw_node) |
| 177 | |
| 178 | #define inet_twsk_for_each_inmate(tw, node, jail) \ |
| 179 | hlist_for_each_entry(tw, node, jail, tw_death_node) |
| 180 | |
| 181 | #define inet_twsk_for_each_inmate_safe(tw, node, safe, jail) \ |
| 182 | hlist_for_each_entry_safe(tw, node, safe, jail, tw_death_node) |
| 183 | |
| 184 | static inline struct inet_timewait_sock *inet_twsk(const struct sock *sk) |
| 185 | { |
| 186 | return (struct inet_timewait_sock *)sk; |
| 187 | } |
| 188 | |
| 189 | static inline u32 inet_rcv_saddr(const struct sock *sk) |
| 190 | { |
| 191 | return likely(sk->sk_state != TCP_TIME_WAIT) ? |
| 192 | inet_sk(sk)->rcv_saddr : inet_twsk(sk)->tw_rcv_saddr; |
| 193 | } |
| 194 | |
| 195 | static inline void inet_twsk_put(struct inet_timewait_sock *tw) |
| 196 | { |
| 197 | if (atomic_dec_and_test(&tw->tw_refcnt)) { |
Arnaldo Carvalho de Melo | eeb2b85 | 2005-10-10 21:25:23 -0700 | [diff] [blame] | 198 | struct module *owner = tw->tw_prot->owner; |
Arnaldo Carvalho de Melo | 8feaf0c | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 199 | #ifdef SOCK_REFCNT_DEBUG |
| 200 | printk(KERN_DEBUG "%s timewait_sock %p released\n", |
| 201 | tw->tw_prot->name, tw); |
| 202 | #endif |
Arnaldo Carvalho de Melo | 6d6ee43 | 2005-12-13 23:25:19 -0800 | [diff] [blame] | 203 | kmem_cache_free(tw->tw_prot->twsk_prot->twsk_slab, tw); |
Arnaldo Carvalho de Melo | eeb2b85 | 2005-10-10 21:25:23 -0700 | [diff] [blame] | 204 | module_put(owner); |
Arnaldo Carvalho de Melo | 8feaf0c | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 205 | } |
| 206 | } |
Arnaldo Carvalho de Melo | e48c414 | 2005-08-09 20:09:46 -0700 | [diff] [blame] | 207 | |
Arnaldo Carvalho de Melo | c676270 | 2005-08-09 20:09:59 -0700 | [diff] [blame] | 208 | extern struct inet_timewait_sock *inet_twsk_alloc(const struct sock *sk, |
| 209 | const int state); |
| 210 | |
Arnaldo Carvalho de Melo | e48c414 | 2005-08-09 20:09:46 -0700 | [diff] [blame] | 211 | extern void __inet_twsk_kill(struct inet_timewait_sock *tw, |
| 212 | struct inet_hashinfo *hashinfo); |
| 213 | |
| 214 | extern void __inet_twsk_hashdance(struct inet_timewait_sock *tw, |
| 215 | struct sock *sk, |
| 216 | struct inet_hashinfo *hashinfo); |
Arnaldo Carvalho de Melo | 696ab2d | 2005-08-09 20:45:03 -0700 | [diff] [blame] | 217 | |
| 218 | extern void inet_twsk_schedule(struct inet_timewait_sock *tw, |
| 219 | struct inet_timewait_death_row *twdr, |
| 220 | const int timeo, const int timewait_len); |
| 221 | extern void inet_twsk_deschedule(struct inet_timewait_sock *tw, |
| 222 | struct inet_timewait_death_row *twdr); |
Arnaldo Carvalho de Melo | 8feaf0c | 2005-08-09 20:09:30 -0700 | [diff] [blame] | 223 | #endif /* _INET_TIMEWAIT_SOCK_ */ |